Everything you want to know about making a video but were afraid to ask.

Know your exact product specs. Decide whether you want reality or an idealized version. Pick one photo as the ground truth.

Product reference photography from multiple angles and reference video for any specific motion will help. Screen specs, audio intent, brand guidelines, and on-screen text requirements help too.

Specific enough to execute. Say what should happen, who it is for, and what it must not feel like.

A useful brief can be as direct as: everyday support for kids, not a sports film; active and healthy, not elderly-coded. The clearer the intention, the more useful the first visual plan becomes.

A good place to begin is a reverse brief so we can find out what you really need.

Often that means building a library of images and flexible video segments that can be used in different ways, depending on whether you are explaining a product, launching a service, or building a brand.

Sometimes. For AI generation, a storyboard can promise a frame-by-frame result that the final process cannot reproduce exactly.

A mood board or a generated sample scene can be a more honest first step. It shows the direction without pretending every frame is already locked.

Maybe, but it's not gonna be very good.

There is a difference between AI-generated video and AI making the whole video. I use AI to help with ideas, scripts, prompts, and organisation.

But I do the shaping. I do the why. I do the interrogation. I do the creative poking and prodding.

What AI cannot add is experience, wisdom, life stories, or the judgment to see the stronger story underneath the brief.

That is a different process. It usually requires 3D modelling, or shooting and lighting the object in a real room.

We can fake it, but it will not have the same ability to complete a full 360 without warping. A slower 180-degree shift is often the more honest compromise.

We can work from phone photos, schematics, reference footage, or very little at all.

A lot of the time, we are visualising things that do not exist yet. You would be amazed what we can do from very, very little.

That is typically done with a 3D model, with several layers composed and animated frame by frame by a team of artists.

It is possible with generative media, but complicated layers moving at once will currently look subpar. The compromise may be to construct the product layer by layer until it is complete.

Because a logo rotating in motion, with changing light and angles, may need to be corrected frame by frame in post.

That is the kind of detail that can take hours per segment. It is Apple-level precision, so we decide together whether it belongs across the whole piece or only in specific shots.

We tested it so you do not have to. Image models reinterpret brand marks: shifted colors, redrawn shapes, and wobbled letterforms.

Real marks come from supplied vector files, comped in post or locked as exact references. A generated logo is a liability wearing a costume.

The first attempt did not get it right. The fix was to stop describing the product in text and let the strongest reference photo do the work, with a short instruction to match it exactly.

Real product interfaces, CAD files, and critical geometry stay protected. AI builds the scene around the hero; it does not get trusted to invent the hero.

One document is declared authoritative. Every retired number is logged on a dead-figure list so it cannot re-enter the copy or prompts.

No new figure enters the film without client sign-off. Boring. Bulletproof.

Every candidate image runs through a 12-point judge: brief compliance, container accuracy, branding, battery modules, camera, lighting, environment, scale, materials, post flags, anti-patterns, and output specification.

The verdicts are APPROVED, APPROVED WITH POST, NEEDS RE-GEN, or FAIL. Lighting comes first, then product accuracy, composition, people, and small details.

No. We scan the existing generation pool first, then re-cut, reframe, and only then re-generate.

Small, precise, or typographic changes often belong in post from the start. The aim is to fix the work without creating new problems elsewhere in the frame.

It happened. In one project, a client caught visible pronation (ankles rolling inward) in two walking shots during the delivery review. The footage showed the movement the product was designed to correct.

One shot was swapped for an alternate from the existing generation pool and another was fixed in the edit. Neutral-gait language was added to future walking prompts.

Technical accuracy is not the same as a point of view. A photorealistic photograph can still feel generic and bland.

We lead with atmosphere and emotion, then build the technical detail into the visual world. The light is part of the story.